TMS320C6657: vlib_c66x_3_3_2_0 VLIB_haarDetectObjectsDense Pedestrian detection

Part Number: TMS320C6747

We are trying to use VLIB 3_3_2_0 to implement a Pedestrian Detection system. And we found there are new APIs for Haar object detection inside the library.

We see the example of the API in the code. However what we want to know for the next step is how we can create our own classifier(VLIB_HAARDETOBJ_sClassifier) to use the APIs. how can I use this function to load a classifier trained with opencv to implement pedestrian detection? If there is an example, please send me a copy.

Part Number: TMS320C6657

hi,

I want to konw how to call the classifier generated by opencv through the VLIB_HAARDETOBJ_sClassifier API？

thanks

Part Number: TMS320C6657

Hi,

I am now trying to implement pedestrian detection using ti's vlib library, but when I try to call the VLIB_haarDetectObjectsDense() function, I don't know how to output the processed image information.This is VLIB_haarDetectObjectsDense（）function,

VLIB_haarDetectObjectsDense(intImageA,
intImageNA,
(prm[tpi].integralImgWidth - classifier->patchSize[1]),
prm[tpi].integralImgPitch,
(prm[tpi].integralImgHeight - classifier->patchSize[0]),
prm[tpi].startStage,
prm[tpi].endStage,
prm[tpi].maxObjects,
1,
classifier,
outObjList,
scratch1,
scratch2);

I hope you can give me some help.
Thanks.

Part Number: TMS320C6657

Hi,

I want to know some formal parameters about VLIB_haarDetectObjectsDense() in vlib_3_3_0_3
Function Documentation question
Int32_t VLIB_haarDetectObjectsDense ( uint32_t integralImgAlign[restrict],
Uint32_t integralImgNonAlign[restrict],
Uint32_t integralImgWidth,
Uint32_t integralImgPitch,
Uint32_t integralImgHeight,
Uint32_t startStage,
Uint32_t endStage,
Uint32_t maxObjects,
Uint32_t yDirJump,
VLIB_HAARDETOBJ_sClassifier classifier[restrict],
VLIB_HAARDETOBJ_sObjectList objListOut[restrict],
Uint8_t scratch1[restrict],
Uint8_t scratch2[restrict]
)
Description:
This function implements object detection based on Haar features with Ada-boost classifier.
Parameters:
[in] integralImgAlign Pointer to 8 byte aligned integral image buffer (UQ32.0)
[in] integralImgNonAlign Pointer to 4 byte (but not 8 byte) aligned integral image, it can be NULL also ? useful for optimization with data in L2 (UQ32.0)
[in] integralImgWidth Width of integral image (UQ32.0)
[in] integralImgPitch Pitch of integral image (UQ32.0)
[in] integralImgHeight Height of integral image (UQ32.0)
[in] startStage classifier stage number to start the dense searching (UQ32.0)
[in] endStage classifier stage number to end the dense processing (UQ32.0)
[in] maxObjects Maximum number of objects to output (raster order FCFS) (UQ32.0)
[in] yDirJump Jump in 'y' co-ordinate direction while search (UQ32.0)
[in] classifier Classifier Data Structure (VLIB_HAARDETOBJ_sClassifier)
[out] objListOut Output List holding the co-ordinates for objects, where each position is given by Y|X in packed format, each being 16 bit quantity (VLIB_HAARDETOBJ_sObjectList)
[in] scratch1 Scratch buffer of size 16 * (integralImgWidth + 4) (UQ8.0)
[in] scratch2 Scratch buffer of size 4 * ((integralImgWidth + 7)/8) (UQ8.0)

How does objListOut pass the parameters, and what information is obtained by scratch1 and scratch2?

thanks.