1. The background

1.1 High-precision data collection

The high precision acquisition vehicle is a mobile surveying and mapping system integrating surveying laser, high performance inertial navigation, high resolution camera and other sensors. After years of deep cultivation, the collection car built by autonavi high precision team has the characteristics of high precision, fast speed, short data generation cycle, high degree of automation, high security, large amount of information and so on.

In order to ensure the precision of high-precision map making, in the high-precision collection car, we use the most advanced laser rangefinder in the industry at present, with the advantages of long measurement distance, point cloud density, scanning frequency can reach 1 million points per second.

1.2 Laser MTA problem

High-speed scanning frequency brings high quality data, but also introduces some special noise and interference, MTA is one of them. What is **MTA (multi-time-around) **? We can see in Figure 1 below. By comparing the two figures, we can see that the MTA problem is actually the ranging problem of the laser. The laser pulls the distant point to the nearby place by mistake, resulting in the distant building becoming the noise on the nearby road.

MTA problems will bring great difficulties to subsequent data processing, automatic identification, map making and other technological processes, resulting in errors in identification and manual processes.

We need to use the laser’s internal mechanisms and data processing algorithms to restore the noise to its original position. This paper will introduce how high precision data processing solves this problem from three aspects: the principle of MTA problem, the internal mechanism of laser to deal with MTA, and the data processing algorithm.

Figure 1 MTA problem data

2. The MTA principle

So how did MTA come about? This starts from the measurement principle of laser.

2.1 Principle of laser ranging

Typical laser scanner is measured by the principle of **TOF (Time of flight) **, that is, the laser sensor emits a pulse every fixed time during measurement, and then measures the returned pulse energy, and calculates the distance of the point according to the time difference between transmitting and receiving:

By periodically “transmitting laser and receiving echo”, the distance of a series of measuring points can be obtained according to the optical flight parameters, and the position of reflecting points can be calculated by combining the position and attitude of laser itself.

2.2 MTA multiple intervals

The laser is limited by its own power, usually can detect the farthest object distance is limited, Dmax. While the emission interval of the laser pulse is DT, the furthest distance that the current laser pulse can detect before the next pulse is:

The laser frequency used by the high-precision acquisition vehicle is 1 million dots/SEC, and the corresponding Dpluse is 150m.

Under normal circumstances, the laser is sent and received in order, that is, send – receive – send – receive, there is always only one laser pulse in the air, receive and send are matched one by one.

However, when Dmax is greater than Dpluse, if the measured object is far away, multiple pulses may appear in the air, and the sequence of multiple pulses arriving at the receiver is no longer consistent with the sequence of pulse transmission. Therefore, the receiver cannot correctly calculate the TOF of the pulse, and thus cannot correctly obtain the location of the object. This is the MULTI-time-around (MTA), as shown in Figure 2 below.

Generally, the number of transmit and receive cycles that the reflected signal may cross is called “MTA interval”. The closest transmitting signal in the matching time is MTA1, and the next-closest transmitting signal is MTA2… And so on.

Dpluse is the interval length of each MTA. If the distance of the object from the laser exceeds this length, the MTA problem will occur. The MTA interval length of the laser of the high precision acquisition vehicle is 150 meters, so the MTA phenomenon will occur for distant buildings that exceed 150 meters.

FIG. 2 MTA interval

3. Internal mechanism of laser response to MTA

In order to deal with the MTA problem, laser manufacturers have made some efforts to find some solutions by using the assumption of surface continuity of the measurement object and variable period measurement technology.

3.1 Neighborhood continuity hypothesis

Most objects in the real world, such as roads, signs, buildings and other man-made objects, have surface continuity and generally do not have drastic geometric changes and textures. Therefore, continuous laser pulse ranging should vary little, as shown in Figure 3.

If a way can be found so that when the laser ranging is misplaced in the MTA region, adjacent laser points no longer have the characteristics of continuity, the point cloud can be placed in the correct MTA region. Variable period measurement technology is based on this idea.

FIG. 3 Continuity of ranging between adjacent laser points

3.2 Variable period measurement technology

In order to identify MTA problems, laser manufacturers have designed a series of patented technologies, the core of which is “variable laser emission interval”, that is, the time interval between the emission of adjacent laser pulses is different, as shown in Figure 4. Moreover, the change of the launch interval is periodic, and its periodic characteristics are shown in Figure 5. When the point cloud is misplaced in the MTA interval, its range is no longer continuous, but jumps back and forth as shown in column 3 of the list in Figure 5. FIG. 6, wrong MTA interval, adjacent points jump back and forth, forming the stratification in the figure.

Figure 4. Variable period emission technology

FIG. 5 Variable period parameters

Figure 6 Error MTA interval

4.MTA correction algorithm

According to the principle of MTA problem and the assumption of neighborhood continuity, combined with the hardware variable period measurement technology, the solution of MTA problem is determined. Firstly, neighborhood division is carried out to find the adjacent laser points, and then the statistical weights of the adjacent points are calculated and put into different MTA intervals. The larger weight is the real MTA interval. In order to improve the performance of the algorithm, the installation position parameters of laser are used to avoid unnecessary weight calculation.

4.1 Neighborhood setting and detection

The neighborhood is determined first, because Lidar scans in circles, taking into account both the contiguity of points that are continuous in time and the contiguity of continuous circles. The basic idea is as follows:

  • Data circle: a circle (line) as the basic processing unit;

  • Continuity calculation region: For a point, adjacent points of its current circle and adjacent points of two adjacent circles are taken as continuity calculation region, as shown in Figure 7.

  • For each point, the range continuity weight and reflectance continuity weight are calculated, i.e. inversely proportional to the variance, and then the MTA region is obtained.

Figure 7. Neighborhood lookup

4.2 Weighted statistical Strategy

The overall weighting strategy is that the larger the distance variance is, the smaller the weight is. The larger the reflectivity variance, the smaller the weight. Gaussian function or trigonometric function is used to select the specific weight.

After statistical analysis of a large number of actual data, the weight of distance variance adopts Gaussian function, where U =0 and δ=0.25, and the weight of reflectivity variance also adopts Gaussian function, where u=0 and δ=4

FIG. 8 Selection of weighting function

The specific calculation process is as follows:

  • For each point, the measurement data as MTA1 and MTA2 are obtained, mainly ranging value and reflectance.

  • For each point, the set of MTA1 and MTA2 neighbor points is obtained.

  • The ranging weights and reflectance weights of each neighbor of each point are calculated, then summed up, and finally the MTA region is determined according to the weights.

4.3 Treatment effect

The processing effect of the algorithm is shown in FIG. 9 and FIG. 10.

Figure 9 MTA treatment effect: untreated MTA

Figure 10 MTA processing effect: MTA recovery result

4.4 Performance Optimization

The basic processing scheme can recover THE MTA error problem well, but the efficiency is very low due to the large search interval and must be processed point by point, which cannot meet the requirements of efficiency and needs to be optimized. The optimization direction considered includes reducing search interval and algorithm optimization.

4.4.1 Reduce the search interval

The detection range parameters of the laser equipment we use are as follows: less than 300m, that is, two MTA intervals. Therefore, we can only consider the two possibilities of MTA1 and MTA2 intervals, which greatly reduces the amount of calculation.

Equipment detection range parameters:

  • The maximum detection range is 235m (80% high reflectivity);

  • Low reflectivity objects less than 100m;

  • Coniferous forest 100m;

  • Asphalt 120m;

  • Broad-leaved forest 150m;

  • Building bricks 200m or so;

  • 250m of white plaster.

4.4.2 Algorithm optimization

The algorithm is further optimized according to the scanning characteristics.

  • Considering that all MTA errors occur above the ground, that is, distant objects that can be actually scanned by the laser are above the ground, points near the ground can be removed according to the vehicle height information.

  • For multiple echoes, the continuity of points is calculated only from the first echo.

  • After dividing the circle, the adjacent space is judged by scanning Angle and ranging value;

  • Split processing, multi-thread parallel acceleration;

  • The points with poor continuity in different regions were removed as isolated points.

5. Summary and outlook

As a part of point cloud solution module, MTA processing algorithm is an important part of cloud data collection and processing. Without solving MTA problem, data collection and processing can not be automated. At the same time, MTA algorithm eliminates the dependence of data processing on laser manufacturer’s software and saves a lot of cost for the company.

In the algorithm design stage, SVM, RF and other machine learning methods were used to solve the problem according to the point cloud classification idea. In the preliminary test, some problems were found, such as difficulty in sample making and large difference between positive and negative sample sizes. On the other hand, machine learning method batch processing needs to consider the appropriate space range, for each block at a rate of 100 million points, its processing efficiency will not meet the needs of the production line.

In the algorithm performance evaluation phase, it was intended to use the results of vendor processing as truth values. However, after evaluation, it is found that the effect of the manufacturer’s processing results is not as good as the self-developed algorithm and can not be used as the evaluation truth value. Finally, combined with the production line process requirements, we specially made an evaluation program, and the algorithm focused on the business requirements, so as to objectively, reliably and quickly complete the evaluation of the algorithm.

At present, the MTA processing algorithm has entered the online production and processed tens of thousands of kilometers of point cloud data. At present, it runs stably and meets expectations.

About the high precision map business center

Autonavi maps is one of its most innovative businesses, dedicated to measuring the world with sensors, understanding the world with algorithms, and redefining the world with data. We cover almost the hottest frontier disciplines, high precision mapping and autonomous driving are multi-disciplinary applied engineering systems. Automatic generation of high precision digital 3D map based on perceptual understanding, 3D reconstruction, fusion positioning, computational geometry technology. Using edge computing, big data processing, cloud services, real-time mass data map reconstruction. Through 5G/V2X information exchange, data exchange between map objects is realized to build a living map. We are not only data producers, but also the definers of a new life. Join us and the future is yours.